tgsi/ureg: add support for output array declarations

4 files changed, 78 insertions, 31 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index b1aebfa1cb9..76ffe9f79a2 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -123,8 +123,11 @@ struct ureg_program
       unsigned semantic_name;
       unsigned semantic_index;
       unsigned usage_mask; /* = TGSI_WRITEMASK_* */
+      unsigned first;
+      unsigned last;
+      unsigned array_id;
    } output[UREG_MAX_OUTPUT];
-   unsigned nr_outputs;
+   unsigned nr_outputs, nr_output_regs;
 
    struct {
       union {
@@ -332,10 +335,12 @@ ureg_DECL_system_value(struct ureg_program *ureg,
 
 
 struct ureg_dst 
-ureg_DECL_output_masked( struct ureg_program *ureg,
-                         unsigned name,
-                         unsigned index,
-                         unsigned usage_mask )
+ureg_DECL_output_masked(struct ureg_program *ureg,
+                        unsigned name,
+                        unsigned index,
+                        unsigned usage_mask,
+                        unsigned array_id,
+                        unsigned array_size)
 {
    unsigned i;
 
@@ -343,7 +348,8 @@ ureg_DECL_output_masked( struct ureg_program *ureg,
 
    for (i = 0; i < ureg->nr_outputs; i++) {
       if (ureg->output[i].semantic_name == name &&
-          ureg->output[i].semantic_index == index) { 
+          ureg->output[i].semantic_index == index) {
+         assert(ureg->output[i].array_id == array_id);
          ureg->output[i].usage_mask |= usage_mask;
          goto out;
       }
@@ -353,6 +359,10 @@ ureg_DECL_output_masked( struct ureg_program *ureg,
       ureg->output[i].semantic_name = name;
       ureg->output[i].semantic_index = index;
       ureg->output[i].usage_mask = usage_mask;
+      ureg->output[i].first = ureg->nr_output_regs;
+      ureg->output[i].last = ureg->nr_output_regs + array_size - 1;
+      ureg->output[i].array_id = array_id;
+      ureg->nr_output_regs += array_size;
       ureg->nr_outputs++;
    }
    else {
@@ -360,16 +370,30 @@ ureg_DECL_output_masked( struct ureg_program *ureg,
    }
 
 out:
-   return ureg_dst_register( TGSI_FILE_OUTPUT, i );
+   return ureg_dst_array_register(TGSI_FILE_OUTPUT, ureg->output[i].first,
+                                  array_id);
 }
 
 
 struct ureg_dst 
-ureg_DECL_output( struct ureg_program *ureg,
-                  unsigned name,
-                  unsigned index )
+ureg_DECL_output(struct ureg_program *ureg,
+                 unsigned name,
+                 unsigned index)
+{
+   return ureg_DECL_output_masked(ureg, name, index, TGSI_WRITEMASK_XYZW,
+                                  0, 1);
+}
+
+struct ureg_dst
+ureg_DECL_output_array(struct ureg_program *ureg,
+                       unsigned semantic_name,
+                       unsigned semantic_index,
+                       unsigned array_id,
+                       unsigned array_size)
 {
-   return ureg_DECL_output_masked(ureg, name, index, TGSI_WRITEMASK_XYZW);
+   return ureg_DECL_output_masked(ureg, semantic_name, semantic_index,
+                                  TGSI_WRITEMASK_XYZW,
+                                  array_id, array_size);
 }
 
 
@@ -1516,10 +1540,12 @@ static void emit_decls( struct ureg_program *ureg )
    for (i = 0; i < ureg->nr_outputs; i++) {
       emit_decl_semantic(ureg,
                          TGSI_FILE_OUTPUT,
-                         i, i,
+                         ureg->output[i].first,
+                         ureg->output[i].last,
                          ureg->output[i].semantic_name,
                          ureg->output[i].semantic_index,
-                         ureg->output[i].usage_mask, 0);
+                         ureg->output[i].usage_mask,
+                         ureg->output[i].array_id);
    }
 
    for (i = 0; i < ureg->nr_samplers; i++) {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index e6304697938..e20f96d5674 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -222,15 +222,24 @@ ureg_DECL_system_value(struct ureg_program *,
                        unsigned semantic_index);
 
 struct ureg_dst
-ureg_DECL_output_masked( struct ureg_program *,
-                         unsigned semantic_name,
-                         unsigned semantic_index,
-                         unsigned usage_mask );
+ureg_DECL_output_masked(struct ureg_program *,
+                        unsigned semantic_name,
+                        unsigned semantic_index,
+                        unsigned usage_mask,
+                        unsigned array_id,
+                        unsigned array_size);
 
 struct ureg_dst
-ureg_DECL_output( struct ureg_program *,
-                  unsigned semantic_name,
-                  unsigned semantic_index );
+ureg_DECL_output(struct ureg_program *,
+                 unsigned semantic_name,
+                 unsigned semantic_index);
+
+struct ureg_dst
+ureg_DECL_output_array(struct ureg_program *ureg,
+                       unsigned semantic_name,
+                       unsigned semantic_index,
+                       unsigned array_id,
+                       unsigned array_size);
 
 struct ureg_src
 ureg_DECL_immediate( struct ureg_program *,
@@ -1175,14 +1184,14 @@ ureg_src_array_offset(struct ureg_src reg, int offset)
 static INLINE struct ureg_dst
 ureg_dst_array_offset( struct ureg_dst reg, int offset )
 {
-   assert(reg.File == TGSI_FILE_TEMPORARY);
    reg.Index += offset;
    return reg;
 }
 
 static INLINE struct ureg_dst
-ureg_dst_register( unsigned file,
-                   unsigned index )
+ureg_dst_array_register(unsigned file,
+                        unsigned index,
+                        unsigned array_id)
 {
    struct ureg_dst dst;
 
@@ -1206,12 +1215,19 @@ ureg_dst_register( unsigned file,
    dst.DimIndFile = TGSI_FILE_NULL;
    dst.DimIndIndex = 0;
    dst.DimIndSwizzle = 0;
-   dst.ArrayID = 0;
+   dst.ArrayID = array_id;
 
    return dst;
 }
 
 static INLINE struct ureg_dst
+ureg_dst_register(unsigned file,
+                  unsigned index)
+{
+   return ureg_dst_array_register(file, index, 0);
+}
+
+static INLINE struct ureg_dst
 ureg_dst( struct ureg_src src )
 {
    struct ureg_dst dst;
diff --git a/src/gallium/state_trackers/nine/nine_ff.c b/src/gallium/state_trackers/nine/nine_ff.c
index e6f2b21dd4d..c2213e6bf11 100644
--- a/src/gallium/state_trackers/nine/nine_ff.c
+++ b/src/gallium/state_trackers/nine/nine_ff.c
@@ -422,13 +422,15 @@ nine_ff_build_vs(struct NineDevice9 *device, struct vs_build_ctx *vs)
     oCol[1] = ureg_saturate(ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1));
 
     if (key->vertexpointsize || key->pointscale) {
-        oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0, TGSI_WRITEMASK_X);
+        oPsz = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_PSIZE, 0,
+                                       TGSI_WRITEMASK_X, 0, 1);
         oPsz = ureg_writemask(oPsz, TGSI_WRITEMASK_X);
     }
     if (key->fog_mode) {
         /* We apply fog to the vertex colors, oFog is for programmable shaders only ?
          */
-        oFog = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_FOG, 0, TGSI_WRITEMASK_X);
+        oFog = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_FOG, 0,
+                                       TGSI_WRITEMASK_X, 0, 1);
         oFog = ureg_writemask(oFog, TGSI_WRITEMASK_X);
     }
 
diff --git a/src/gallium/state_trackers/nine/nine_shader.c b/src/gallium/state_trackers/nine/nine_shader.c
index cdd4918530b..22a58825f78 100644
--- a/src/gallium/state_trackers/nine/nine_shader.c
+++ b/src/gallium/state_trackers/nine/nine_shader.c
@@ -1098,7 +1098,7 @@ _tx_dst_param(struct shader_translator *tx, const struct sm1_dst_param *param)
         if (ureg_dst_is_undef(tx->regs.oDepth))
            tx->regs.oDepth =
               ureg_DECL_output_masked(tx->ureg, TGSI_SEMANTIC_POSITION, 0,
-                                      TGSI_WRITEMASK_Z);
+                                      TGSI_WRITEMASK_Z, 0, 1);
         dst = tx->regs.oDepth; /* XXX: must write .z component */
         break;
     case D3DSPR_PREDICATE:
@@ -1966,7 +1966,7 @@ DECL_SPECIAL(DCL)
                 tx->info->position_t = TRUE;
             assert(sem.reg.idx < Elements(tx->regs.o));
             tx->regs.o[sem.reg.idx] = ureg_DECL_output_masked(
-                ureg, tgsi.Name, tgsi.Index, sem.reg.mask);
+                ureg, tgsi.Name, tgsi.Index, sem.reg.mask, 0, 1);
 
             if (tgsi.Name == TGSI_SEMANTIC_PSIZE)
                 tx->regs.oPts = tx->regs.o[sem.reg.idx];
@@ -1984,7 +1984,8 @@ DECL_SPECIAL(DCL)
         if (!is_input && 0) { /* declare in COLOROUT/DEPTHOUT case */
             /* FragColor or FragDepth */
             assert(sem.reg.mask != 0);
-            ureg_DECL_output_masked(ureg, tgsi.Name, tgsi.Index, sem.reg.mask);
+            ureg_DECL_output_masked(ureg, tgsi.Name, tgsi.Index, sem.reg.mask,
+                                    0, 1);
         }
     }
     return D3D_OK;
@@ -2312,7 +2313,8 @@ DECL_SPECIAL(TEXM3x2DEPTH)
     ureg_CMP(ureg, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_negate(ureg_abs(ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y))),
              ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X), ureg_imm1f(ureg, 1.0f));
     /* replace the depth for depth testing with the result */
-    tx->regs.oDepth = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_POSITION, 0, TGSI_WRITEMASK_Z);
+    tx->regs.oDepth = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_POSITION, 0,
+                                              TGSI_WRITEMASK_Z, 0, 1);
     ureg_MOV(ureg, tx->regs.oDepth, ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
     /* note that we write nothing to the destination, since it's disallowed to use it afterward */
     return D3D_OK;
@@ -2410,7 +2412,8 @@ DECL_SPECIAL(TEXDEPTH)
     ureg_CMP(ureg, ureg_writemask(r5, TGSI_WRITEMASK_X), ureg_negate(ureg_abs(r5g)),
              r5r, ureg_imm1f(ureg, 1.0f));
     /* replace the depth for depth testing with the result */
-    tx->regs.oDepth = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_POSITION, 0, TGSI_WRITEMASK_Z);
+    tx->regs.oDepth = ureg_DECL_output_masked(ureg, TGSI_SEMANTIC_POSITION, 0,
+                                              TGSI_WRITEMASK_Z, 0, 1);
     ureg_MOV(ureg, tx->regs.oDepth, r5r);
 
     return D3D_OK;